Strategy
- Start small — begin with test jobs to determine actual resource needs
- Match resources to requirements — request only the cores and memory your job will use
- Consider wait times — smaller requests often queue faster
- Use shared queues — for jobs that don't need a full node
- Verify compatibility — ensure software works with the target CPU/GPU architecture
Note: Memory only needs to be specified for shared queues; otherwise all available node memory is requested by default.
Legacy queues (login1/login2) — 28-core Intel Haswell
AVX2 support, 128 GB memory per node.
| Queue | Duration | Max nodes | Shared | Best for |
|---|---|---|---|---|
debug-28core | 1 hr | 8 | No | Testing/debugging |
short-28core | 1–4 hr | 12 | No | Small/medium jobs |
medium-28core | 4–12 hr | 24 | No | Medium parallel jobs |
long-28core | 8–48 hr | 8 | No | Long simulations |
extended-28core | 8 hr – 7 days | 2 | No | Very long simulations |
large-28core | 4–8 hr | 80 | No | Large parallel jobs |
Legacy GPU queues
| Queue | GPU | GPU memory | Duration | Best for |
|---|---|---|---|---|
gpu | 4× K80 | 24 GB | 1–8 hr | Basic GPU workloads |
gpu-long | 4× K80 | 24 GB | 8–48 hr | Long GPU jobs |
p100 | 2× P100 | 16 GB | 1–24 hr | Scientific computing |
v100 | 2× V100 | 32 GB | 1–24 hr | ML/AI |
Modern queues (milan1/milan2) — 40-core Intel Skylake
AVX512 support, 192 GB memory per node.
| Queue | Duration | Max nodes | Shared | Best for |
|---|---|---|---|---|
debug-40core | 1 hr | 8 | No | Testing/debugging |
short-40core | 1–4 hr | 8 | No | Standard jobs |
short-40core-shared | 1–4 hr | 4 | Yes | Smaller jobs |
medium-40core | 4–12 hr | 16 | No | Medium jobs |
long-40core | 8–48 hr | 6 | No | Long jobs |
long-40core-shared | 8–24 hr | 3 | Yes | Shared long jobs |
extended-40core | 8 hr – 7 days | 2 | No | Very long jobs |
extended-40core-shared | 8 hr – 3.5 days | 1 | Yes | Shared extended jobs |
large-40core | 4–8 hr | 50 | No | Large parallel jobs |
96-core AMD EPYC Milan
AVX2 support, 256 GB memory per node.
| Queue | Duration | Max nodes | Shared | Best for |
|---|---|---|---|---|
short-96core | 1–4 hr | 8 | No | Parallel jobs |
short-96core-shared | 1–4 hr | 4 | Yes | Moderate shared jobs |
medium-96core | 4–12 hr | 16 | No | Parameter sweeps |
long-96core | 8–48 hr | 6 | No | Long parallel jobs |
long-96core-shared | 8–24 hr | 3 | Yes | Shared long jobs |
extended-96core | 8 hr – 7 days | 2 | No | Very long jobs |
extended-96core-shared | 8 hr – 3.5 days | 1 | Yes | Shared extended jobs |
large-96core | 4–8 hr | 38 | No | Large parallel jobs |
High-bandwidth memory (HBM) nodes
AMX/AVX512 support, 384 GB per node (256 GB DDR5 + 128 GB HBM). Use HBM nodes for workloads limited by memory bandwidth.
| Queue | Duration | Max nodes | Special features | Best for |
|---|---|---|---|---|
hbm-short-96core | 1–4 hr | 8 | High-bandwidth memory | Memory-intensive jobs |
hbm-medium-96core | 4–12 hr | 16 | Enhanced memory performance | Large datasets |
hbm-long-96core | 8–48 hr | 6 | 2–4× memory speed | Memory-bound simulations |
hbm-extended-96core | 8 hr – 7 days | 2 | Maximum duration | Long memory jobs |
hbm-large-96core | 4–8 hr | 38 | Large-scale HBM computing | Massive memory-bound jobs |
hbm-1tb-long-96core | 8–48 hr | 1 | 1 TB memory + 128 GB HBM cache | Extremely large datasets |
NVIDIA A100 GPU nodes
4× A100 80 GB GPUs on 64-core Intel Ice Lake nodes (256 GB memory). Use GPU queues only for compatible applications — verify CUDA/software requirements before submission.
| Queue | Duration | Max nodes | Shared | Best for |
|---|---|---|---|---|
a100 | 1–8 hr | 2 | Yes | GPU workloads, AI/ML |
a100-long | 8–48 hr | 1 | Yes | Long GPU jobs |
a100-large | 1–8 hr | 4 | Yes | Large GPU jobs |
Applies to